Triage: v0.39.0 — a global first bound in a top-level for body is no longer visible outside the loop (if-block bindings still are)

[InauguralPhysicist]

Noticed downstream in ouroboros (harness hardening, ouroboros#101/#102): the parity program module_name_in_block.eigs had silently degraded — its second half now errors identically on both compilers, which the old harness masked as a green string-compare. Needs an upstream verdict: intended for-scope semantics, or regression?

Repro at v0.39.0:

for q in [1, 2]:
    acc is q
print of acc

Error line 3: undefined variable 'acc'

Same through a function read (define show() as: return acc after the loop) — undefined. The asymmetry is the suspicious part: an if-block first-binding IS still a module binding:

if 1 == 1:
    c is 7
define g() as:
    return c
print of (g of null)     # prints 7

So at v0.39.0 a name first bound in a top-level for body is loop-local (invisible even at module level after the loop), while if/other block bindings remain module bindings. Unknown which release changed it — the downstream harness vacuity hid the transition (any pin bump since the program was written is possible).

If intended: say so here and document the for-scope rule; ouroboros keeps the matched-behavior reject case (added in ouroboros#102) until told otherwise.
If regression: the ouroboros case is a matched-bug canary and will fail the moment a fixed pin restores the old behavior — that failure is the reminder to flip the case back to full parity in the same EIGS_REF bump.

🤖 Generated with Claude Code